The LLM benchmark aggregator.

DeepSeek-V4-Pro (High, 20260813)

DeepSeek · 1.7T · open weights

Strongest in Cybersecurity (#15 of 46), weakest in Toxicity avoidance (#214 of 271). Above par in 25 of 33 scopes. Among the models it meets almost everywhere, it finishes behind Claude Fable 5 and Claude Fable 5.1 and ahead of GPT-5.6 Luna and GPT-5.

§ 1 · Profile

What it is good at

Bars run from 50 — the average of the models each source lists — so right of the line is above par. The middle column is the gap to whoever leads that scope.

Coding58.7−7.7#17/1762/6

Agentic coding59.1−7.9#16/1692/5

Code generation47.4−17.9—/20✱0/1

Repository Q&A50.5−16.2—/0✱✱

Agents56.8−9.1#32/2892/6

Knowledge work60.9−10.9#38/1982/2

Workflow automation44.7−17.3—/0✱✱

Reasoning54.9−10#47/1904/5

Science61.4−2.1#17/1221/1

Reasoning52.7−14#53/1442/3

Mathematics53.7−10.4#66/1502/2

Expert reasoning46.5−14.4—/0✱✱

Human preference63−5.6#49/3501/1

Professional52.3−10.8#63/1881/2

Cybersecurity57.2−6.8#15/461/1

IT operations40.6−33.2#25/311/1

Finance55.2−7.3#45/1611/1

Legal53.2−12.7#62/1641/1

Medical50.8−15.7#79/1491/1

Knowledge53−9.7#63/1911/3

Academic knowledge57.2−6.7#35/1231/1

Core abilities53−12.2#72/2253/4

Language55.3−11.3#24/621/1

Instruction following49.5−24.3#33/621/1

General intelligence53.8−15#72/2142/3

Safety50.4−11#166/3312/3

Factual grounding53.5−17.3#42/1021/1

Jailbreak resistance54.4−10.6#123/2711/1

Harm refusal51.4−10.1#132/2961/2

Secure code49.7−16.6#157/2731/1

Fairness45.2−24.8#188/2961/2

Toxicity avoidance48−11#214/2711/1

§ 2 · Head to head

What it beats, and what beats it

The same models turn up scope after scope. Counted once: where both were placed, who finished higher. Bars run right for DeepSeek-V4-Pro (High, 20260813), left for the other.

Mistral Large 3won 20 · lost 0

Gemini 3.5 Flash-Litewon 21 · lost 0

DeepSeek V4 Flashwon 30 · lost 1

DeepSeek V3won 24 · lost 2

Grok 4.3won 22 · lost 2

Command A (03-2025)won 21 · lost 2

GPT-5.4 miniwon 19 · lost 2

GPT-5.4 nanowon 19 · lost 3

Llama 3.3 70B Instructwon 17 · lost 3

Kimi K2won 20 · lost 4

NVIDIA-Nemotron-3-Ultra-550B-A55B (NVFP4)won 24 · lost 5

Llama 4 Scout Instructwon 19 · lost 4

Grok 4.1 Fastwon 18 · lost 4

GPT OSS 20Bwon 17 · lost 4

Kimi K2.6won 20 · lost 5

GPT-4.1 nanowon 19 · lost 5

GPT-4.1 miniwon 19 · lost 5

GLM-4.6won 19 · lost 5

Gemma 4 31Bwon 19 · lost 5

Mistral Large 4won 18 · lost 5

Gemini 2.0 Flashwon 17 · lost 5

MiMo V2.5 Prowon 16 · lost 5

Llama 4 Maverick Instructwon 19 · lost 6

GPT OSS 120Bwon 19 · lost 6

Claude Haiku 4.5won 19 · lost 6

DeepSeek R1won 19 · lost 6

DeepSeek V3.2won 19 · lost 6

Kimi K2.5won 19 · lost 6

Claude 3.5 Haikuwon 15 · lost 5

GPT-4o miniwon 18 · lost 6

Grok 3won 17 · lost 6

Qwen3.8 27Bwon 17 · lost 6

Gemini 2.5 Flashwon 18 · lost 7

MiniMax M3won 21 · lost 10

GPT-4owon 16 · lost 8

Qwen3 235B A22B Instructwon 16 · lost 8

Claude Sonnet 4won 17 · lost 9

Claude 3.5 Sonnetwon 13 · lost 7

MiniMax M2.7won 14 · lost 8

GLM-5.1won 14 · lost 8

GLM-5.2won 19 · lost 11

Claude 3.7 Sonnetwon 13 · lost 8

GPT-6 Lunawon 16 · lost 10

Gemini 2.5 Prowon 16 · lost 10

Inklingwon 19 · lost 12

O3 Miniwon 14 · lost 9

Grok 4won 14 · lost 9

O1won 12 · lost 8

Qwen3.7 Maxwon 15 · lost 10

GPT-5 nanowon 13 · lost 9

GLM-5.3 Flashwon 16 · lost 12

GPT-5.1won 13 · lost 10

O4 Miniwon 14 · lost 11

Claude Sonnet 4.5 (High, 32K, 20250929)won 14 · lost 11

Grok 3 Miniwon 11 · lost 9

MiMo V2.6 Flashwon 12 · lost 10

GPT-4.1won 13 · lost 11

Claude Opus 4won 13 · lost 11

GPT-5 miniwon 13 · lost 12

GPT-5won 13 · lost 12

GPT-5.6 Lunawon 16 · lost 15

Kimi K2 Thinkingwon 10 · lost 10

GPT-5.2won 13 · lost 13

Claude Sonnet 4.6won 12 · lost 13

Gemini 3.6 Flashwon 12 · lost 14

O3won 10 · lost 14

Gemini 3.1 Prowon 9 · lost 13

Muse Spark 1.2won 8 · lost 12

Grok 4.7won 10 · lost 15

Gemini 3.5 Flashwon 10 · lost 15

GLM-5.3won 10 · lost 15

Muse Spark 1.1won 9 · lost 15

Grok 4.5won 10 · lost 18

Claude Sonnet 5won 11 · lost 20

Claude Opus 4.5 (High, 32K, 20251101)won 7 · lost 14

Gemini 3.8 Flashwon 9 · lost 19

Claude Opus 4.6won 7 · lost 16

Qwen3.8 Maxwon 8 · lost 19

Claude Opus 4.7won 8 · lost 19

Gemini 3 Prowon 6 · lost 15

Gemini 3.7 Flashwon 6 · lost 16

MiMo V2.6 Prowon 6 · lost 17

GPT-6 Solwon 7 · lost 20

GPT-5.4won 7 · lost 20

GPT-5.6 Terrawon 8 · lost 23

GPT-5.5won 7 · lost 21

Grok 4.6won 6 · lost 21

Kimi K3won 5 · lost 19

GPT-6 Astrawon 4 · lost 18

DeepSeek V4.1 Flashwon 4 · lost 21

GPT-5.6 Solwon 5 · lost 27

GPT-6.1 Solwon 3 · lost 17

Claude Opus 4.8won 3 · lost 23

Claude Opus 5won 3 · lost 28

Claude Opus 5.5won 2 · lost 24

Claude Sonnet 5.5won 1 · lost 19

Claude Fable 5.1won 1 · lost 21

Claude Fable 5won 0 · lost 20

§ 3 · Sources

Where the numbers come from

12 publications, 53 figures. Every one links to the page it was read from.

LMArena Text ↗1 measureread 2026-10-21

LMArena Text 1465

Artificial Analysis ↗1 measureread 2026-10-21

Artificial Analysis Intelligence Index 36

GDPval-AA ↗1 measureread 2026-10-21

GDPval-AA 1449

AA-Briefcase ↗1 measureread 2026-10-21

AA-Briefcase 1256

ARC-AGI-2 ↗1 measureread 2026-10-21

ARC-AGI-2 61.3%

LiveBench 1.0 ↗8 measuresread 2026-10-21

LiveBench 1.0 77.4LiveBench 1.0 · Reasoning 85.8LiveBench 1.0 · Coding 77.2LiveBench 1.0 · Agentic Coding 54.9LiveBench 1.0 · Mathematics 95.1LiveBench 1.0 · Data Analysis 79.2LiveBench 1.0 · Language 82.1LiveBench 1.0 · Instruction Following 67.7

Kagi LLM Benchmark ↗1 measureread 2026-10-21

Kagi LLM Benchmark 53.5%

SimpleBench ↗1 measureread 2026-10-21

SimpleBench 50.9%

Vals.ai ↗16 measuresread 2026-10-21

Vals · Legal Research Bench 40.87%Vals · LegalBench 82.36%Vals · Harvey Legal Agent Benchmark 10.83%Vals · Finance Agent 50.39%Vals · CorpFin 65.42%Vals · TaxEval 73.06%Vals · MedCode 42.47%Vals · MedScribe 80.17%Vals · CyberBench 72.86%Vals · SRE Bench 1.91%Vals · SWE-bench Verified 96.4% (Mini-SWE-agent)Vals · Vibe Code Bench 82.3% (OpenHands)Vals · Code Migration 41.54%Vals · GPQA Diamond 92.42%Vals · MMLU Pro 86.97%Vals · ProofBench 50%

Enkrypt AI Safety Leaderboard ↗6 measuresread 2026-10-21

Enkrypt · Jailbreak risk 9.1%Enkrypt · Harmful content risk 22.2%Enkrypt · CBRN risk 17.3%Enkrypt · Toxicity risk 7.5%Enkrypt · Bias risk 85.3%Enkrypt · Insecure code risk 33.8%

Vectara Hallucination Leaderboard ↗1 measureread 2026-10-21

Vectara · Factual consistency 91.4%

✱✱ DeepSeek — V4.1 Flash model card ↗the model’s own publisher15 measuresread 2026-10-21

GPQA Diamond (Pass@1) 92.4%Codeforces (Rating) 3348MathArena Apex (Pass@1) 65.3%Terminal-Bench 2.1 (Pass@1) 87.9%Terminal-Bench 3.0 (Pass@1) 11.8%Terminal-Bench 4.0 (Pass@1) 12.4%DeepSWE v1.1 (Resolved) 62.7%ProgramBench (Almost@1) 15.5%NL2Repo-Bench (Score) 61.5%CyberGym (Pass@1) 83.3%SEC-Bench Pro (Pass@1) 56.4%ExploitGym (Pass@1) 5.4%HLE w/ tools (Pass@1) 60%AutomationBench (Pass@1) 43.2%Agent's Last Exam (Pass@1) 25.7%

Badge

[![PublicAI Index](https://publicai.io/model-index/badge?model=deepseek-v4-pro-high-20260813)](https://publicai.io/model-index/m/deepseek-v4-pro-high-20260813)